# API changes: October 25, 2026

[← Changelog](/docs/changelog) · [October 2026](/docs/changelog/2026-10)

### NEW-1025-1: preview and connect models to a key of any provider

[GET /v1/ai/credentials/:id/available-models](/docs/ai/credentials/available-models) returns the provider's live list and each model's connection state without changing key models. If listing is unavailable, the response remains HTTP 200 with empty `models` and `unavailable.code: provider_list_models_unavailable`.

[POST /v1/ai/credentials/:id/models](/docs/ai/credentials/models-add) connects models of any provider. A single request preserves HTTP 201; a duplicate returns `409 already_exists`. The new batch `{ modelIds: [...] }` of up to 100 identifiers returns HTTP 200 with `created` and `skipped` arrays; existing connections are marked `already_connected`. An invalid batch is rejected in full before writing. One key holds at most 1000 models: exceeding this returns `409 key_models_limit`.

[GET /v1/ai/credentials/:id/models](/docs/ai/credentials/models-list) additionally returns `capabilities`, `maxOutputTokens` and `description`.